News leaders expect search traffic to fall 43%.

The search referral is shrinking

For twenty years, news publishers lived on a simple bargain: search engines sent readers, publishers sold their attention. AI answers are breaking it. Google search traffic to more than 2,500 news sites fell 33% between November 2024 and November 2025, and Google Discover fell 21%, according to Chartbeat data in the Reuters Institute trends report. Media leaders expect search referrals to fall by another 43% over the next three years.

−33% Google search traffic to 2,500+ news sites, Nov 2024 to Nov 2025 (Chartbeat via Reuters Institute)

The survey behind that expectation covers 280 media leaders in 51 countries, the Reuters Institute release says. Two thirds of them report that AI efficiencies have not yet saved any jobs. Up to one in five expect substantial revenue from AI platforms in 2026. The industry is betting on AI and bracing for it at the same time.

That double position shapes buying. A newsroom that licenses its archive to an AI platform on Monday may be negotiating with the same company's answer engine about attribution on Tuesday. Before a single tool is chosen, three questions need an owner: which content is used to train or ground AI systems, on what terms, and how the brand appears when an AI answer quotes it. The tools themselves come second. A transcription service, a versioning assistant or a dubbing engine is easy to pilot; the rights position behind it is not. Media buyers who settle the rights first move faster later, because every pilot then starts from known terms.

Audiences are moving to answers

The other side of the bargain is the reader. In the Reuters Institute Digital News Report 2026, 10% of people across 45 markets use AI chatbots for news each week, up from 7% a year earlier, The Decoder summarises. Among 18- to 24-year-olds it is 17%. Only 1% call a chatbot their main source, but the habit is forming in the youngest group first.

The problem for publishers is what happens after the answer. Only 4% of chatbot news users always or often click through to the original source, against 19% for search engines and 17% for social media. A chatbot answer can carry a publisher's reporting without carrying its reader.

Source: Reuters Institute Digital News Report 2026, 27 markets; as summarised by The Decoder, 19 Jun 2026.

Trust lags use. 20% of people trust news from AI chatbots, compared with 37% who trust most news. Among chatbot users the figure is higher, which is exactly why accuracy and attribution in AI answers matter to the brands being summarised.

Licensing becomes the business model

If traffic falls, the value of an archive, a character library or a newsroom's daily output shifts to licensing. The largest signal so far came from entertainment. On 11 December 2025 Disney announced a $1 billion investment in OpenAI and a licence for more than 200 animated characters, with costumes and props, for Sora and ChatGPT, ABC News reported.

US$1bn Disney's investment in OpenAI, with a character licence for Sora (Dec 2025)

The same studio litigates where it has no deal. In June 2025 Disney and Universal sued the image generator Midjourney in a California federal court, alleging that it trained on their characters and reproduces them. The complaint cites more than 150 works and seeks up to $150,000 per infringed work; it describes Midjourney as having 21 million subscribers and $300 million in revenue the year before.

Courts have not given a single answer. In the UK, the High Court ruled on 4 November 2025 in Getty Images v Stability AI. Getty had dropped its primary copyright claims during trial; the secondary infringement claim was dismissed, and Getty won only on some trademark points that the judge called extremely limited in scope, as Cleary Gottlieb summarises. For media buyers the lesson is practical. The law will take years to settle. Contracts, licences and provenance records decide your exposure now.

Video, creators and the new front page

Publishers are also changing what they make. 76% of leaders in the Reuters Institute survey will push staff to act more like creators in 2026, and half plan to partner with creators to distribute content. Planned extra investment is strongest on YouTube, ahead of TikTok and Instagram.

AI is part of that shift: transcription, versioning, translation, clipping and synthetic voice make one story available in more formats and languages at lower cost. The Good AI Prism places these tools across several petals: Media & Entertainment for newsroom and studio workflows, Design, Image & Video for generation and editing, and Meetings, Voice & Language for transcription, dubbing and translation. In the Good AI Prism Global Market Study, the media row names two outcomes to measure: cost per content unit and licensing revenue.

Labels are now law

Synthetic media needs a label almost everywhere. Under Article 50 of the EU AI Act, applying since 2 August 2026, deployers who publish deepfakes must disclose that the content was artificially generated or manipulated. AI-generated text published to inform the public on matters of public interest must also be disclosed, unless it has gone through human review and a person holds editorial responsibility. Generative systems already on the market have until 2 December 2026 to mark output in a machine-readable format, under the Digital Omnibus on AI.

China's labelling measures have applied since 1 September 2025, with explicit labels and metadata on text, images, audio and video. South Korea's AI Basic Act has applied since 22 January 2026. Our synthetic media rules analysis compares the regimes in detail.

The exception for human editorial control matters for newsrooms. It rewards exactly what good journalism already does: a named editor who checks and answers for what is published.

In practice

Best practice: a named editor on every AI-assisted piece. The AI Act exempts AI-generated text on matters of public interest from the disclosure duty when it has passed human review and a person holds editorial responsibility. The routine that earns the exemption is the one good newsrooms already run: a named editor reviews, approves and answers for each AI-assisted piece, and the CMS records who it was.

Scenario: an archive licence offer. An AI platform offers a regional publisher a fee for its archive. Before signing, the publisher settles three points: which content is in and which stays out, whether answers attribute and link to the source, and what happens to the licence if the platform's answer engine cuts search referrals further. Its own click-through data is the baseline for that negotiation.

What to do Monday

1. Measure the AI referral. Track traffic from AI assistants and answer engines separately from search, and set a baseline before you change anything.
2. Audit every generative tool. List each tool that produces text, image, video or voice, with its licence terms, training-data statement and how it marks output for 2 December 2026.
3. Write rights into contracts. Require indemnities, provenance records and consent for any cloned voice or likeness, including staff and presenters.
4. Price your archive. Decide which content you license to AI platforms, on what terms, and which you protect; record what is already being used.
5. Keep an editor accountable. Put a named editor on every AI-assisted publication. It is good journalism and, in the EU, the condition for the text exception.